8The Urbit interpreter is built on a Nock runtime system written
9in C, `u3`. This document is a relatively complete description
11You should keep reading if (a) you're planning to work on the
12Urbit interpreter; (b) you're a language implementation geek; or
13(c) you don't really understand anything until you've seen the
16## u3: noun processing in C
18`u3` is the C library that makes Urbit work. If it wasn't called
19`u3`, it might be called `libnoun` - it's a library for making
22What's a noun? A noun is either a cell or an atom. A cell is an
23ordered pair of any two nouns. An atom is an unsigned integer of
26To the C programmer, this is not a terribly complicated data
27structure, so why do you need a library for it?
29One: nouns have a well-defined computation kernel, Nock, whose
30spec fits on a page and gzips to 340 bytes. But the only
31arithmetic operation in Nock is increment. So it's nontrivial
32to compute both efficiently and correctly.
34Two: `u3` is designed to be a "solid-state interpreter," ie, a
35single-level store which is transparently snapshotted. This
36implies a specialized memory-management model, etc, etc.
38(Does `u3` depend on the higher levels of Urbit, Arvo and Hoon?
39Yes and no. `u3` expects you to load something shaped like an
40Arvo kernel, and use it as an event-processing function. But you
41don't need to use this feature if you don't want, and your kernel
42doesn't have to be Arvo proper - just Arvo-compatible. Think of
43`u3` as the BIOS and Arvo as the boot kernel. And there are no
44dependencies at all between Hoon the language and `u3`.)
48Under `u3` is the simple `c3` layer, which is just how we write C
51When writing C in u3, please of course follow the conventions of
52the code around you as regards indentation, etc. It's especially
53important that every function have a header comment, even if it
54says nothing interesting.
56But some of our idiosyncrasies go beyond convention. Yes, we've
57done awful things to C. Here's what we did and why we did.
61First, it's generally acknowledged that underspecified integer
62types are C's worst disaster. C99 fixed this, but the `stdint`
63types are wordy and annoying. We've replaced them with:
67 typedef uint64_t c3_d; // double-word
68 typedef int64_t c3_ds; // signed double-word
69 typedef uint32_t c3_w; // word
70 typedef int32_t c3_ws; // signed word
71 typedef uint16_t c3_s; // short
72 typedef int16_t c3_ss; // signed short
73 typedef uint8_t c3_y; // byte
74 typedef int8_t c3_ys; // signed byte
75 typedef uint8_t c3_b; // bit
77 typedef uint8_t c3_t; // boolean
78 typedef uint8_t c3_o; // loobean
79 typedef uint8_t c3_g; // 5-bit atom for a 32-bit log.
80 typedef uint32_t c3_l; // little; 31-bit unsigned integer
81 typedef uint32_t c3_m; // mote; also c3_l; LSB first a-z 4-char string.
85 typedef char c3_c; // does not match int8_t or uint8_t
86 typedef int c3_i; // int - really bad
87 typedef uintptr_t c3_p; // pointer-length uint - really really bad
88 typedef intptr_t c3_ps; // pointer-length int - really really bad
90Some of these need explanation. A loobean is a Nock boolean -
91Nock, for mysterious reasons, uses 0 as true (always say "yes")
92and 1 as false (always say "no").
94Nock and/or Hoon cannot tell the difference between a short atom
95and a long one, but at the `u3` level every atom under `2^31` is
96direct. The `c3_l` type is useful to annotate this. A `c3_m` is
97a *mote* - a string of up to 4 characters in a `c3_l`, least
98significant byte first. A `c3_g` should be a 5-bit atom. Of
99course, C cannot enforce these constraints, only document them.
101Use the "bad" - ie, poorly specified - integer types only when
102interfacing with external code that expects them.
104An enormous number of motes are defined in `i/c/motes.h`. There
105is no reason to delete motes that aren't being used, or even to
106modularize the definitions. Keep them alphabetical, though.
108### c3: variables and variable naming
110The C3 style uses Hoon style TLV variable names, with a quasi
111Hungarian syntax. This is weird, but works really well, as long
112as what you're doing isn't hideously complicated. (Then it works
113badly, but we shouldn't need anything hideous in u3.)
115A TLV variable name is a random pronounceable three-letter
116string, sometimes with some vague relationship to its meaning,
117but usually not. Usually CVC (consonant-vowel-consonant) is a
120You should use TLVs much the way math people use Greek letters.
121The same concept should in general get the same name across
122different contexts. When you're working in a given area, you'll
123tend to remember the binding from TLV to concept by sheer power
124of associative memory. When you come back to it, it's not that
125hard to relearn. And of course, when in doubt, comment it.
127Variables take pseudo-Hungarian suffixes, matching in general the
128suffix of the integer type:
130 c3_w wor_w; // 32-bit word
132Unlike in standard Hungarian, there is no change for pointer
133variables. C structure variables take a `_u` suffix.
137The code (from `defs.h`) tells the story:
142 # define _(x) (c3y == (x))
143 # define __(x) ((x) ? c3y : c3n)
144 # define c3a(x, y) __(_(x) && _(y))
145 # define c3o(x, y) __(_(x) || _(y))
147In short, use `_()` to turn a loobean into a boolean, `__` to go
148the other way. Use `!` as usual, `c3y` for yes and `c3n` for no,
149`c3a` for and and `c3o` for or.
153The division between `c3` and `u3` is that you could theoretically
154imagine using `c3` as just a generic C environment. Anything to do
155with nouns is in `u3`.
157### u3: a map of the system
159There are two kinds of symbols in `u3`: regular and irregular.
160Regular symbols follow this pattern:
163 -------------------------------------------------------
164 u3a_ allocation i/n/a.h n/a.c
165 u3e_ persistence i/n/e.h n/e.c
166 u3h_ hashtables i/n/h.h n/h.c
167 u3i_ noun construction i/n/i.h n/i.c
168 u3j_ jet control i/n/j.h n/j.c
169 u3m_ system management i/n/m.h n/m.c
170 u3n_ nock computation i/n/n.h n/n.c
171 u3r_ noun access (error returns) i/n/r.h n/r.c
172 u3t_ profiling i/n/t.h n/t.c
173 u3v_ arvo i/n/v.h n/v.c
174 u3x_ noun access (error crashes) i/n/x.h n/x.c
175 u3z_ memoization i/n/z.h n/z.c
176 u3k[a-g] jets (transfer, C args) i/j/k.h j/[a-g]/*.c
177 u3q[a-g] jets (retain, C args) i/j/q.h j/[a-g]/*.c
178 u3w[a-g] jets (retain, nock core) i/j/w.h j/[a-g]/*.c
180Irregular symbols always start with `u3` and obey no other rules.
181They're defined in `i/n/u.h`. Finally, `i/all.h` includes all
182these headers (fast compilers, yay) and is all you need to
185### u3: noun internals
187A noun is a `u3_noun` - currently defined as a 32-bit `c3_w`.
189If your `u3_noun` is less than `(1 << 31)`, it's a direct atom.
190Every unsigned integer between `0` and `0x7fffffff` inclusive is
193If bit `31` is set in a `u3_noun`, bit `30` is always set - this
194bit is reserved. Bit `29` is `1` if the noun is a cell, `0` if
195it's an atom. Bits `28` through `0` are a word pointer into the
196loom - see below. The structures are:
201 c3_w buf_w[0]; // actually [len_w]
210The only thing that should be mysterious here is `mug_w`, which
211is a 31-bit lazily computed nonzero short hash (FNV currently,
212soon Murmur3). If `mug_w` is 0, the hash is not yet computed.
213We also hijack this field for various hacks, such as saving the
214new address of a noun when copying over.
216Also, the value `0xffffffff` is `u3_none`, which is never a valid
217noun. Use the type `u3_weak` to express that a noun variable may
220### u3: reference counts
222The only really essential thing you need to know about `u3` is
223how to handle reference counts. Everything else, you can skip
226u3 deals with reference-counted, immutable, acyclic nouns.
227Unfortunately, we are not Apple and can't build reference
228counting into your C compiler, so you need to count by hand.
230Every allocated noun (or any allocation object, because our
231allocator is general-purpose) contains a counter which counts the
232number of references to it - typically variables with type
233`u3_noun`. When this counter goes to 0, the noun is freed.
235To tell `u3` that you've added a reference to a noun, call the
236function `u3a_gain()` or its shorthand `u3k()`. (For your
237convenience, this function returns its argument.) To tell `u3`
238that you've destroyed a reference, call `u3a_lose()` or `u3z()`.
240(If you screw up by decrementing the counter too much, `u3` will
241dump core in horrible ways. If you screw up by incrementing it
242too much, `u3` will leak memory. To check for memory leaks,
243set the `bug_o` flag in `u3e_boot()` - eg, run `vere` with `-g`.
244Memory leaks are difficult to debug - the best way to handle
245leaks is just to revert to a version that didn't have them, and
246look over your code again.)
248(You can gain or lose a direct atom. It does nothing.)
250### u3: reference protocols
252*THIS IS THE MOST CRITICAL SECTION IN THE `u3` DOCUMENTATION.*
254The key question when calling a C function in a refcounted world
255is what the function will do to the noun refcounts - and, if the
256function returns a noun, what it does to the return.
258There are two semantic patterns, `transfer` and `retain`. In
259`transfer` semantics, the caller "gives" a use count to the
260callee, which "gives back" any return. For instance, if I have
263 u3_noun foo = u3i_string("foobar");
271Suppose `u3f_futz()` has `transfer` semantics. At `[...]`, my
272code holds one reference to `bar` and zero references to `foo` -
273which has been freed, unless it's part of `bar`. My code now
274owns `bar` and gets to work with it until it's done, at which
275point a `u3z()` is required.
277On the other hand, if `u3f_futz()` has `retain` semantics, we
281 u3_noun foo = u3i_string("foobar");
289because calling `u3f_futz()` does not release our ownership of
290`foo`, which we have to free ourselves.
292But if we free `bar`, we are making a great mistake, because our
293reference to it is not in any way registered in the memory
294manager (which cannot track references in C variables, of
295course). It is normal and healthy to have these uncounted
296C references, but they must be treated with care.
298The bottom line is that it's essential for the caller to know
299the refcount semantics of any function which takes or returns a
300noun. (In some unusual circumstances, different arguments or
301returns in one function may be handled differently.)
303Broadly speaking, as a design question, retain semantics are more
304appropriate for functions which inspect or query nouns. For
305instance, `u3h()` (which takes the head of a noun) retains, so
306that we can traverse a noun tree without constantly incrementing
309Transfer semantics are more appropriate for functions which make
310nouns, which is obviously what most functions do.
312In general, though, in most places it's not worth thinking about
313what your function does. There is a convention for it, which
314depends on where it is, not what it does. Follow the convention.
316### u3: reference conventions
318The `u3` convention is that, unless otherwise specified, *all
319functions have transfer semantics* - with the exception of the
320prefixes: `u3r`, `u3x`, `u3z`, `u3q` and `u3w`. Also, within
321jet directories `a` through `f` (but not `g`), internal functions
322retain (for historical reasons).
324If functions outside this set have retain semantics, they need to
325be commented, both in the `.h` and `.c` file, with `RETAIN` in
326all caps. Yes, it's this important.
328### u3: system architecture
330If you just want to tinker with some existing code, it might be
331enough to understand the above. If not, it's probably worth
332taking the time to look at `u3` as a whole.
334`u3` is designed to work as a persistent event processor.
335Logically, it computes a function of the form
337 f(event, old state) -> (actions, new state)
339Obviously almost any computing model - including, but not limited
340to, Urbit - can be defined in this form. To create the illusion
341of a computer that never loses state and never fails, we:
343- log every event externally before it goes into u3
344- keep a single reference to a permanent state noun.
345- can abort any event without damaging the permanent state.
346- snapshot the permanent state periodically, and/or prune logs.
348### u3: the road model
350`u3` uses a memory design which I'm sure someone has invented
351somewhere before, because it's not very clever, but I've never
352seen it anywhere in particular.
354Every allocation starts with a solid block of memory, which `u3`
355calls the `loom`. How do we allocate on the loom? You're
356probably familiar with the Unix heap-stack design, in which the
357stack grows downward and the heap (malloc arena) grows upward:
361 |------------#################################+++++++++++++|
365A road is a normal heap-stack system, except that the heap
366and stack can point in *either direction*. Therefore, inside
367a road, we can nest another road in the *opposite direction*.
369When the opposite road completes, its heap is left on top of
370the opposite heap's stack. It's no more than the normal
371behavior of a stack machine for all subcomputations to push
372their results on the stack.
374The performance tradeoff of "leaping" - reversing directions in
375the road - is that if the outer computation wants to preserve the
376results of the inner one, not just use them for temporary
377purposes, it has to *copy them*.
379This is a trivial cost in some cases, a prohibitive cost in
380others. The upside, of course, is that all garbage accrued
381in the inner computation is discarded at zero cost.
383The goal of the road system is the ability to *layer* memory
384models. If you are allocating on a road, you have no idea
385how deep within a nested road system you are - in other words,
386you have no idea exactly how durable your result may be.
387But free space is never fragmented within a road.
389Roads do not reduce the generality or performance of a memory
390system, since even the most complex GC system can be nested
391within a road at no particular loss of performance - a road
392is just a block of memory.
394Each road (`u3a_road` to be exact) uses four pointers: `rut` is
395the bottom of the arena, `hat` the top of the arena, `mat` the
396bottom of the stack, `cap` the top of the stack. (Bear in mind
397that the road "stack" is not actually used as the C function-call
398stack, though it probably should be.)
400A "north" road has the stack high and the heap low:
404 |~~~~~~~~~~~~-------##########################+++++++$~~~~~|
408A "south" road is the other way around:
412 |~~~~~~~~~~~~$++++++##########################--------~~~~~|
416Legend: `-` is durable storage (heap); `+` is temporary storage
417(stack); `~` is deep storage (immutable); `$` is the allocation
418frame `#` is free memory.
420Pointer restrictions: pointers stored in `+` can point anywhere.
421Pointers in `-` can only point to `-` or `~`; pointers in `~`
424To "leap" is to create a new inner road in the `###` free space.
425but in the reverse direction, so that when the inner road
426"falls" (terminates), its durable storage is left on the
427temporary storage of the outer road.
429`u3` keeps a global variable, `u3_Road` or its alias `u3R`, which
430points to the current road. (If we ever run threads in inner
431roads - see below - this will become a thread-local variable.)
432Relative to `u3R`, `+` memory is called `junior` memory; `-`
433memory is `normal` memory; `~` is `senior` memory.
435### u3: explaining the road model
439We're now ready to understand why the road system works so
440logically with the event and persistence model.
442The key is that *we don't update refcounts in senior memory.*
443A pointer from an inner road to an outer road is not counted.
444Also, the outmost, or `surface` road, is the only part of the
445image that gets checkpointed.
447So the surface road contains the entire durable state of `u3`.
448When we process an event, or perform any kind of complicated or
449interesting calculation, *we process it in an inner road*. If
450its results are saved, they need to be copied.
452Since processing in an inner road does not touch surface memory,
453(a) we can leave the surface road in a read-only state and not
454mark its pages dirty; (b) we can abort an inner calculation
455without screwing up the surface; and (c) because inner results
456are copied onto the surface, the surface doesn't get fragmented.
458All of (a), (b) and (c) are needed for checkpointing to be easy.
459It might be tractable otherwise, but easy is even better.
461Moreover, while the surface is most definitely single-threaded,
462we could easily run multiple threads in multiple inner roads
463(as long as the threads don't have pointers into each others'
464memory, which they obviously shouldn't).
466Moreover, in future, we'll experiment more with adding road
467control hints to the programmer's toolbox. Reference counting is
468expensive. We hypothesize that in many - if not most - cases,
469the programmer can identify procedural structures whose garbage
470should be discarded in one step by copying the results. Then,
471within the procedure, we can switch the allocator into `sand`
472mode, and stop tracking references at all.
474### u3: rules for C programming
476There are two levels at which we program in C: (1) above the
477interpreter; (2) within the interpreter or jets. These have
478separate rules which need to be respected.
480### u3: rules above the interpreter
482In its relations with Unix, Urbit follows a strict rule of "call
483me, I won't call you." We do of course call Unix system calls,
484but only for the purpose of actually computing.
486Above Urbit, you are in a normal C/Unix programming environment
487and can call anything in or out of Urbit. Note that when using
488`u3`, you're always on the surface road, which is not thread-safe
489by default. Generally speaking, `u3` is designed to support
490event-oriented, single-threaded programming.
492If you need threads which create nouns, you could use
493`u3m_hate()` and `u3m_love()` to run these threads in subroads.
494You'd need to make the global road pointer, `u3R`, a thread-local
495variable instead. This seems perfectly practical, but we haven't
496done it because we haven't needed to.
498### u3: rules within the interpreter
500Within the interpreter, your code can run either in the surface
501road or in a deep road. You can test this by testing
505ie: does the pier's home road equal the current road pointer?
507Normally in this context you assume you're obeying the rules of
508running on an inner road, ie, "deep memory." Remember, however,
509that the interpreter *can* run on surface memory - but anything
510you can do deep, you can do on the surface. The converse is by
513In deep memory, think of yourself as if in a signal handler.
514Your execution context is extremely fragile and may be terminated
515without warning or cleanup at any time (for instance, by ^C).
517For instance, you can't call `malloc` (or C++ `new`) in your C
518code, because you don't have the right to modify data structures
519at the global level, and will leave them in an inconsistent state
520if your inner road gets terminated. (Instead, use our drop-in
521replacements, `u3a_malloc()`, `u3a_free()`, `u3a_realloc()`.)
523A good example is the different meaning of `c3_assert()` inside
524and outside the interpreter. At either layer, you can use
525regular assert(), which will just kill your process. On the
526surface, `c3_assert()` will just... kill your process.
528In deep execution, `c3_assert()` will issue an exception that
529queues an error event, complete with trace stack, on the Arvo
530event queue. Let's see how this happens.
534You produce an exception with
536 /* u3m_bail(): bail out. Does not return.
540 ** %exit :: semantic failure
541 ** %evil :: bad crypto
542 ** %intr :: interrupt
543 ** %fail :: execution failure
544 ** %foul :: assert failure
545 ** %need :: network block
546 ** %meme :: out of memory
547 ** %time :: timed out
548 ** %oops :: assertion failure
551 u3m_bail(c3_m how_m);
553Broadly speaking, there are two classes of exception: internal
554and external. An external exception begins in a Unix signal
555handler. An internal exception begins with a call to longjmp()
558There are also two kinds of exception: mild and severe. An
559external exception is always severe. An internal exception is
560normally mild, but some (like `c3__oops`, generated by
561`c3_assert()`) are severe.
563Either way, exceptions come with a stack trace. The `u3` nock
564interpreter is instrumented to retain stack trace hints and
565produce them as a printable `(list tank)`.
567Mild exceptions are caught by the first virtualization layer and
568returned to the caller, following the behavior of the Nock
569virtualizer `++mock` (in `hoon.hoon`)
571Severe exceptions, or mild exceptions at the surface, terminate
572the entire execution stack at any depth and send the cumulative
573trace back to the `u3` caller.
575For instance, `vere` uses this trace to construct a `%crud`
576event, which conveys our trace back toward the Arvo context where
577it crashed. This lets any UI component anywhere, even on a
578remote node, render the stacktrace as a consequence of the user's
579action - even if its its direct cause was (for instance) a Unix
582### u3: C structures on the loom
584Normally, all data on the loom is nouns. Sometimes we break this
585rule just a little, though - eg, in the `u3h` hashtables.
587To point to non-noun C structs on the loom, we use a `u3_post`,
588which is just a loom word offset. A macro lets us declare this
589as if it was a pointer:
591 typedef c3_w u3_post;
592 #define u3p(type) u3_post
594Some may regard this as clever, others as pointless. Anyway, use
595`u3to()` and `u3of()` to convert to and from pointers.
597When using C structs on the loom - generally a bad idea - make
598sure anything which could be on the surface road is structurally
599portable, eg, won't change size when the pointer size changes.
600(Note also: we consider little-endian, rightly or wrongly, to
601have won the endian wars.)
603## u3: API overview by prefix
605Let's run through the `u3` modules one by one. All public
606functions are commented, but the comments may be cryptic.
612 /* u3m_boot(): start the u3 system.
615 u3m_boot(c3_o nuu_o, c3_o bug_o, c3_c* dir_c);
617`nuu_o` is `c3y` (yes, `0`) if you're creating a new pier,
618`c3n` (no, `1`) if you're loading an existing one. `bug_o`
619is `c3y` if you want to test the garbage-collector, `c3n`
620otherwise. `dir_c` is the directory for the pier files.
622`u3m_boot()` expects an `urbit.pill` file to load the kernel
623from. It will try first `$dir/.urb.urbit.pill`, then `U3_LIB`.
625Any significant computation with nouns, certainly anything Turing
626complete, should be run (a) virtualized and (b) in an inner road.
627These are slightly different things, but at the highest level we
628bundle them together for your convenience, in `u3m_soft()`:
630 /* u3m_soft(): system soft wrapper. unifies unix and nock errors.
632 ** Produces [%$ result] or [%error (list tank)].
635 u3m_soft(c3_w sec_w, u3_funk fun_f, u3_noun arg);
637`sec_w` is the number of seconds to time out the computation.
638`fun_f` is a C function accepting `arg`.
640The result of `u3m_soft()` is a cell whose head is an atom. If
641the head is `%$` - ie, `0` - the tail is the result of
642`fun_f(arg)`. Otherwise, the head is a `term` (an atom which is
643an LSB first string), and the tail is a `(list tank)` (a list of
644`tank` printables - see `++tank` in `hoon.hoon`). Error terms
645should be the same as the exception terms above.
647If you're confident that your computation won't fail, you can
648use `u3m_soft_sure()`, `u3m_soft_slam()`, or `u3m_soft_nock()`
649for C functions, Hoon function calls, and Nock invocations.
650Caution - this returns just the result, and asserts globally.
652All the `u3m_soft` functions above work *only on the surface*.
653Within the surface, virtualize with `u3m_soft_run()`. Note that
654this takes a `fly` (a namespace gate), thus activating the `11`
655super-operator in the nock virtualizer, `++mock`. When actually
656using the `fly`, call `u3m_soft_esc()`. Don't do either unless
657you know what you're doing!
659For descending into a subroad *without* Nock virtualization,
660use `u3m_hate()` and `u3m_love` respectively. Hating enters
661a subroad; loving leaves it, copying out a product noun.
663Other miscellaneous tools in `u3m`: `u3m_file()` loads a Unix
664file as a Nock atom; `u3m_water()` measures the boundaries of
665the loom in current use (ie, watermarks); and a variety of
666prettyprinting routines, none perfect, are available, mainly for
667debugging printfs: `u3m_pretty()`, `u3m_p()`, `u3m_tape()` and
670It's sometimes nice to run a mark-and-sweep garbage collector,
671`u3m_grab()`, which collects the world from a list of roots,
672and asserts if it finds any leaks or incorrect refcounts. This
673tool is for debugging and long-term maintenance only; refcounts
678The jet system, `u3j`, is what makes `u3` and `nock` in any sense
679a useful computing environment. Except perhaps `u3a` (there is
680really no such thing as a trivial allocator, though `u3a` is
681dumber than most) - `u3j` is the most interesting code in `u3`.
683Let's consider the minor miracle of driver-to-battery binding
684which lets `u3j` work - and decrement not be `O(n)` - without
685violating the precisely defined semantics of pure Nock, *ever*.
687It's easy to assume that jets represent an architectural coupling
688between Hoon language semantics and Nock interpreter internals.
689Indeed such a coupling would be wholly wrongtious and un-Urbit.
690But the jet system is not Hoon-specific. It is specific to nock
691runtime systems that use a design pattern we call a `core`.
693#### u3j: core structure
695A core is no more than a cell `[code data]`, in which a `code` is
696either a Nock formula or a cell of `code`s, and `data` is anything.
697In a proper core, the subject each formula expects is the core
700Except for the arbitrary decision to make a core `[code data]`,
701(or as we sometimes say, `[battery payload]`), instead of `[data
702code]`, any high-level language transforming itself to Nock would
705So jets are in fact fully general. Broadly speaking, the jet
706system works by matching a C *driver* to a battery. When the
707battery is invoked with Nock operator `9`, it must be found in
708associative memory and linked to its driver. Then we link the
709formula axis of the operation (`a` in `[9 a b]`) to a specific
710function in the driver.
712To validate this jet binding, we need to know two things. One,
713we need to know the C function actually is a perfect semantic
714match for the Nock formula. This can be developed with driver
715test flags, which work, and locked down with a secure formula
716hash in the driver, which we haven't bothered with just yet.
717(You could also try to develop a formal method for verifying
718that C functions and Nock formulas are equivalent, but this is
719a research problem for the future.)
721Two, we need to validate that the payload is appropriate for the
722battery. We should note that jets are a Nock feature and have no
723reference to Hoon. A driver which relies on the Hoon type system
724to only pair it with valid payloads is a broken driver, and
725breaks the Nock compliance of the system as a whole. So don't.
727Now, a casual observer might look at `[battery payload]` and
728expect the simplest case of it to be `[formula subject]`. That
729is: to execute a simple core whose battery is a single formula,
734Then, naturally, when we go from Hoon or a high-level language
735containing functions down to Nock, `[function arguments]` turns
736into `[formula subject]`. This seems like an obvious design, and
737we mention it only because it is *completely wrong*.
739Rather, to execute a one-armed core like the above, we run
743and the normal structure of a `gate`, which is simply Urbitese
746 [formula [sample context]]
748where `sample` is Urbitese for "arguments" - and `context`, any
749Lisper will at once recognize, is Urbitese for "environment."
751To `slam` or call the gate, we simply replace the default sample
752with the caller's data, then nock the formula on the entire gate.
754What's in the context? Unlike in most dynamic languages, it is
755not some secret system-level bag of tricks. Almost always it is
756another core. This onion continues until at the bottom, there is
757an atomic constant, conventionally is the kernel version number.
759Thus a (highly desirable) `static` core is one of the form
762 [battery static-core]
764ie, a solid stack of nested libraries without any dynamic data.
765The typical gate will thus be, for example,
767 [formula [sample [battery battery battery constant]]]
769but we would be most foolish to restrict the jet mechanism to
770cores of this particular structure. We cannot constrain a
771payload to be `[sample static-core]`, or even `[sample core]`.
772Any such constraint would not be rich enough to handle Hoon,
773let alone other languages.
777There are two fundamental rules of computer science: (1) every
778system is best understood through its state; (2) less state is
779better than more state. Sadly, a pier has three different jet
780state systems: `cold`, `warm` and `hot`. It needs all of them.
782Hot state is associated with this particular Unix process. The
783persistent pier is portable not just between process and process,
784but machine and machine or OS and OS. The set of jets loaded
785into a pier may itself change (in theory, though not in the
786present implementation) during the lifetime of the process. Hot
787state is a pure C data structure.
789Cold state is associated with the logical execution history of
790the pier. It consists entirely of nouns and ignores restarts.
792Warm state contains all dependencies between cold and hot
793state. It consists of C structures allocated on the loom.
795Warm state is purely a function of cold and hot states, and
796we can wipe and regenerate it at any time. On any restart where
797the hot state might have changed, we clear the warm state
800There is only one hot state, the global jet dashboard
801`u3j_Dash` or `u3D` for short. In the present implementation,
802u3D is a static structure not modified at runtime, except for
803numbering itself on process initialization. This structure -
804which embeds function pointers to all the jets - is defined
805in `j/tree.c`. The data structures:
807 /* u3j_harm: driver arm.
809 typedef struct _u3j_harm {
810 c3_c* fcs_c; // `.axe` or name
811 u3_noun (*fun_f)(u3_noun); // compute or 0 / semitransfer
812 c3_o ice; // perfect (don't test)
813 c3_o tot; // total (never punts)
814 c3_o liv; // live (enabled)
817 /* u3j_core: C core driver.
819 typedef struct _u3j_core {
820 c3_c* cos_c; // control string
821 struct _u3j_harm* arm_u; // blank-terminated static list
822 struct _u3j_core* dev_u; // blank-terminated static list
823 struct _u3j_core* par_u; // dynamic parent pointer
824 c3_l jax_l; // dynamic jet index
827 /* u3e_dash, u3_Dash, u3D: jet dashboard singleton
829 typedef struct _u3e_dash {
830 u3j_core* dev_u; // null-terminated static list
831 c3_l len_l; // ray_u filled length
832 c3_l all_l; // ray_u allocated length
833 u3j_core* ray_u; // dynamic driver array
836Warm and cold state is *per road*. In other words, as we nest
837roads, we also nest jet state. The jet state in the road is:
839 struct { // jet dashboard
840 u3p(u3h_root) har_p; // warm state
841 u3_noun das; // cold state
844In case you understand Hoon, `das` (cold state) is a `++dash`,
845and `har_p` (warm state) is a map from battery to `++calx`:
847 ++ bane ,@tas :: battery name
848 ++ bash ,@uvH :: label hash
849 ++ bosh ,@uvH :: local battery hash
850 ++ batt ,* :: battery
852 $: jax=,@ud :: hot core index
853 hap=(map ,@ud ,@ud) :: axis/hot arm index
854 lab=path :: label as path
855 jit=* :: arbitrary data
857 ++ calx (trel calf (pair bash cope) club) :: cached by battery
858 ++ clog (pair cope (map batt club)) :: identity record
859 ++ club (pair corp (map term nock)) :: battery pattern
860 ++ cope (trel bane axis (each bash noun)) :: core pattern
862 ++ corp (each core batt) :: parent or static
863 ++ dash (map bash clog) :: jet system
865The driver index `jax` in a `++calx` is an index into `ray_u` in the
866dashboard - ie, a pointer into hot state. This is why the warm
867state has to be reset when we reload the pier in a new process.
869Why is jet state nested? Nock of course is a functional system,
870so as we compute we don't explicitly create state. Jet state is
871an exception to this principle (which works only because it can't
872be semantically detected from Nock/Hoon) - but it can't violate
873the fundamental rules of the allocation system.
875For instance, when we're on an inner road, we can't allocate on
876an outer road, or point from an outer road to an inner. So if we
877learn something - like a mapping from battery to jet - in the
878inner road, we have to keep it in the inner road.
880Mitigating this problem, when we leave an inner road (with
881`u3m_love()`), we call `u3j_reap()` to promote jet information in
882the dying road. Reaping promotes anything we've learned about
883any battery that either (a) already existed in the outer road, or
884(b) is being saved to the outer road.
888Jet binding starts with a `%fast` hint. (In Hoon, this is
889produced by the runes `~%`, for the general case, or `~/`
890for simple functions.) To bind a jet, execute a formula of the
893 [10 [%fast clue-formula] core-formula]
895`core-formula` assembles the core to be jet-propelled.
896`clue-formula` produces the hint information, or `++clue`
897above, which we want to annotate it with.
899A clue is a triple of name, parent, and hooks:
901 ++ clue (trel chum nock (list (pair term nock)))
903The name, or `++chum`, has a bunch of historical structure which
904we don't need (cleaning these things up is tricky), but just gets
905flattened into a term.
907The parent axis is a nock formula, but always reduces to a simple
908axis, which is the address of this core's *parent*. Consider
909again an ordinary gate
911 [formula [sample context]]
913Typically the `context` is itself a library core, which itself
914has a jet binding. If so, the parent axis of this gate is `7`.
916If the parent is already bound - and the parent *must* be already
917bound, in this road or a road containing it - we can hook this core
918bottom-up into a tree hierarchy. Normally the child core is
919produced by an arm of the parent core, so this is not a problem -
920we wouldn't have the child if we hadn't already made the parent.
922The clue also contains a list of *hooks*, named nock formulas on
923the core. Usually these are arms, but they need not be. The
924point is that we often want to call a core from C, in a situation
925where we have no type or other source information. A common case
926of this is a complex system in which we're mixing functions which
927are jet-propelled with functions that aren't.
929In any case, all the information in the `%fast` hint goes to
930`u3j_mine()`, which registers the battery in cold state (`das` in
931`jed` in `u3R`), then warm state (`har_p` in `jed`).
933It's essential to understand that the `%fast` hint has to be,
934well, fast - because we apply it whenever we build a core. For
935instance, if the core is a Hoon gate - a function - we will call
936`u3j_mine` every time the function is called.
938### u3j: the cold jet dashboard
940For even more fun, the jet tree is not actually a tree of
941batteries. It's a tree of battery *labels*, where a label is
942an [axis term] path from the root of the tree. (At the root,
943if the core pattern is always followed properly, is a core whose
944payload is an atomic constant, conventionally the Hoon version.)
946Under each of these labels, it's normal to have an arbitrary
947number of different Nock batteries (not just multiple copies
948of the same noun, a situation we *do* strive to avoid). For
949instance, one might be compiled with debugging hints, one not.
951We might even have changed the semantics of the battery without
952changing the label - so long as those semantics don't invalidate
955et tree. For instance, it's normal to have
956two equivalent Nock batteries at the same time in one pier: one
957battery compiled with debugging hints, one not.
959Rather, the jet tree is a semantic hierarchy. The root of the
960hierarchy is a constant, by convention the Hoon kernel version
961because any normal jet-propelled core has, at the bottom of its
962onion of libraries, the standard kernel. Thus if the core is
964 [foo-battery [bar-battery [moo-battery 164]]]
966we can reverse the nesting to construct a hierarchical core
971extends the static core `164/moo/bar` by wrapping the `foo`
972battery (ie, in Hoon, `|%`) around it. With the core above,
973you can compute `foo` stuff, `bar` stuff, and `moo` stuff.
976Not all cores are static, of course - they may contain live data,
977like the sample in a gate (ie, argument to a function). Once
978again, it's important to remember that we track jet bindings not
979by the core, which may not be static, but by the battery, which
982(And if you're wondering how we can use a deep noun like a Nock
983formula or battery as a key in a key-value table, remember
984`mug_w`, the lazily computed short hash, in all boxed nouns.)
986In any case, `das`, the dashboard, is a map from `bash` to jet
987location record (`++clog`). A `clog` in turn contains two kinds
988of information: the `++cope`, or per-location noun; and a map of
989batteries to a per-battery `++club`.
991The `cope` is a triple of `++bane` (battery name, right now just
992a `term`); `++axis`, the axis, within *this* core, of the parent;
993and `(each bash noun)`, which is either `[0 bash]` if the parent
994is another core, or `[1 noun]`, for the constant noun (like
995`164`) if there is no parent core.
997A `bash` is just the noun hash (`++sham`) of a `cope`, which
998uniquely expresses the battery's hierarchical location without
999depending on the actual formulas.
1001The `club` contains a `++corp`, which we use to actually validate
1002the core. Obviously jet execution has to be perfectly compatible
1003with Nock. We search on the battery, but getting the battery
1004right is not enough - a typical battery is dependent on its
1005context. For example, your jet-propelled library function is
1006very likely to call `++dec` or other advanced kernel technology.
1007If you've replaced the kernel in your context with something
1008else, we need to detect this and not run the jet.
1010There are two cases for a jet-propelled core - either the entire
1011core is a static constant, or it isn't. Hence the definition
1014 ++ corp (each core batt) :: parent or static
1016Ie, a `corp` is `[0 core]` or `[1 batt]`. If it's static -
1017meaning that the jet only works with one specific core, ie, the
1018parent axis of each location in the hierarchy is `3` - we can
1019validate with a single comparison. Otherwise, we have to recurse
1020upward by checking the parent.
1022Note that there is at present no way to force a jet to depend on
1025### u3j: the warm jet dashboard
1027We don't use the cold state to match jets as we call them. We
1028use the cold state to register jets as we find them, and also to
1029rebuild the warm state after the hot state is reset.
1031What we actually use at runtime is the warm state, `jed->har_p`,
1032which is a `u3h` (built-in hashtable), allocated on the loom,
1033from battery to `++calx`.
1035A `calx` is a triple of a `++calf`, a `[bash cope]` cell, and a
1036`club`. The latter two are all straight from cold state.
1038The `calf` contains warm data dependent on hot state. It's a
1039quadruple: of `jax`, the hot driver index (in `ray_u` in
1040`u3j_dash`); `hap`, a table from arm axis (ie, the axis of each
1041formula within the battery) to driver arm index (into `arm_u` in
1042`u3j_core`); `lab`, the complete label path; and `jit`, any
1043other dynamic data that may speed up execution.
1045We construct `hap`, when we create the calx, by iterating through
1046the arms registered in the `u3j_core`. Note the way a `u3j_harm`
1047declares itself, with the string `fcs_c` which can contain either
1048an axis or a name. Most jetted cores are of course gates, which
1049have one formula at one axis within the core: `fcs_c` is `".3"`.
1051But we do often have fast cores with more complex arm structure,
1052and it would be sad to have to manage their axes by hand. To use
1053an `fcs_c` with a named arm, it's sufficient to make sure the
1054name is bound to a formula `[0 axis]` in the hook table.
1056`jit`, as its name suggests, is a stub where any sort of
1057optimization data computed on battery registration might go. To
1058use it, fill in the `_cj_jit()` function.
1060### u3j: the hot dashboard
1062Now it should be easy to see how we actually invoke jets. Every
1063time we run a nock `9` instruction (pretty often, obviously), we
1064have a core and an axis. We pass these to `u3j_kick()`, which
1065will try to execute them.
1067Because nouns with a reference count of 1 are precious,
1068`u3j_kick()` has a tricky reference control definition. It
1069reserves the right to return `u3_none` in the case where there is
1070no driver, or the driver does not apply for this case; in this
1071case, it retains argument `cor`. If it succeeds, though, it
1074`u3j_kick()` searches for the battery (always the head of the
1075core, of course) in the hot dashboard. If the battery is
1076registered, it searches for the axis in `hap` in the `calx`.
1077If it exists, the core matches a driver and the driver jets this
1078arm. If not, we return `u3_none`.
1080Otherwise, we call `fun_f` in our `u3j_harm`. This obeys the
1081same protocol as `u3j_kick()`; it can refuse to function by
1082returning `u3_none`, or consume the noun.
1084Besides the actual function pointer `fun_f`, we have some flags
1085in the `u3j_harm` which tell us how to call the arm function.
1087If `ice` is yes (`&`, `0`), the jet is known to be perfect and we
1088can just trust the product of `fun_f`. Otherwise, we need to run
1089*both* the Nock arm and `fun_f`, and compare their results.
1091(Note that while executing the C side of this test, we have to
1092set `ice` to yes; on the Nock side, we have to set `liv` to no.
1093Otherwise, many non-exponential functions become exponential.
1094When auto-testing jets in this way, the principle is that the
1095test is on the outermost layer of recursion.)
1097(Note also that anyone who multi-threads this execution
1098environment has a slight locking problem with these flags if arm
1099testing is multi-threaded.)
1101If `tot` is yes, (`&`, `0`), the arm function is *total* and has
1102to return properly (though it can still return *u3_none*).
1103Otherwise, it is *partial* and can `u3_cm_bail()` out with
1104c3__punt. This feature has a cost: the jet runs in a subroad.
1106Finally, if `liv` is no (`|`, 1), the jet is off and doesn't run.
1108It should be easy to see how the tree of cores gets declared -
1109precisely, in `j/dash.c`. We declare the hierarchy as a tree
1110of `u3j_core` structures, each of which comes with a static list
1111of arms `arm_u` and sub-cores `dev_u`.
1113In `u3j_boot()`, we traverse the hierarchy, fill in parent
1114pointers `par_u`, and enumerate all `u3j_core` structures
1115into a single flat array `u3j_dash.ray_u`. Our hot state
1116then appears ready for action.
1118### u3j: jet functions
1120At present, all drivers are compiled statically into `u3`. This is
1121not a long-term permanent solution or anything. However, it will
1122always be the case with a certain amount of core functionality.
1124For instance, there are some jet functions that we need to call
1125as part of loading the Arvo kernel - like `++cue` to unpack a
1126noun from an atom. And obviously it makes sense, when jets are
1127significant enough to compile into `u3`, to export their symbols
1128in headers and the linker.
1130There are three interface prefixes for standard jet functions:
1131`u3k`, `u3q`, and `u3w`. All jets have `u3w` interfaces; most
1132have `u3q`; some have `u3k`. Of course the actual logic is
1135`u3w` interfaces use the same protocol as `fun_f` above: the
1136caller passes the entire core, which is retained if the function
1137returns `u3_none`, transferred otherwise. Why? Again, use
1138counts of 1 are special and precious for performance hackers.
1140`u3q` interfaces break the core into C arguments, *retain* noun
1141arguments, and *transfer* noun returns. `u3k` interfaces are the
1142same, except with more use of `u3_none` and other simple C
1143variations on the Hoon original, but *transfer* both arguments
1144and returns. Generally, `u3k` are most convenient for new code.
1146Following `u3k/q/w` is `[a-f]`, corresponding to the 6 logical
1147tiers of the kernel, or `g` for user-level jets. Another letter
1148is added for functions within subcores. The filename, under
1149`j/`, follows the tier and the function name.
1151For instance, `++add` is `u3wa_add(cor)`, `u3qa_add(a, b)`, or
1152`u3ka_add(a, b)`, in `j/a/add.c`. `++get` in `++by` is
1153`u3wdb_get(cor)`, `u3kdb_get(a, b)`, etc, in `j/d/by_get.c`.
1155For historical reasons, all internal jet code in `j/[a-f]`
1156*retains* noun arguments, and *transfers* noun results. Please
1157do not do this in new `g` jets! The new standard protocol is to
1158transfer both arguments and results.
1160### u3a: allocation functions
1162`u3a` allocates on the current road (u3R). Its internal
1163structures are uninteresting and typical of a naive allocator.
1165The two most-used `u3a` functions are `u3a_gain()` to add a
1166reference count, and `u3a_lose()` to release one (and free the
1167noun, if the use count is zero). For convenience, `u3a_gain()`
1168returns its argument. The pair are generally abbreviated with
1169the macros `u3k()` and `u3z()` respectively.
1171Normally we create nouns through `u3i` functions, and don't call
1172the `u3a` allocators directly. But if you do:
1174One, there are *two* sets of allocators: the word-aligned
1175allocators and the fully-aligned (ie, malloc compatible)
1176allocators. For instance, on a typical OS X setup, malloc
1177produces 16-byte aligned results - needed for some SSE
1180These allocators are *not compatible*. For 32-bit alignment
1181as used in nouns, call
1183 /* u3a_walloc(): allocate storage measured in words.
1186 u3a_walloc(c3_w len_w);
1188 /* u3a_wfree(): free storage.
1191 u3a_wfree(void* lag_v);
1193 /* u3a_wealloc(): word realloc.
1196 u3a_wealloc(void* lag_v, c3_w len_w);
1198For full alignment, call:
1200 /* u3a_malloc(): aligned storage measured in bytes.
1203 u3a_malloc(size_t len_i);
1205 /* u3a_realloc(): aligned realloc in bytes.
1208 u3a_realloc(void* lag_v, size_t len_i);
1210 /* u3a_realloc2(): gmp-shaped realloc.
1213 u3a_realloc2(void* lag_v, size_t old_i, size_t new_i);
1215 /* u3a_free(): free for aligned malloc.
1218 u3a_free(void* tox_v);
1220 /* u3a_free2(): gmp-shaped free.
1223 u3a_free2(void* tox_v, size_t siz_i);
1225There are also a set of special-purpose allocators for building
1226atoms. When building atoms, please remember that it's incorrect
1227to have a high 0 word - the word length in the atom structure
1228must be strictly correct.
1230Of course, we don't always know how large our atom will be.
1231Therefore, the standard way of building large atoms is to
1232allocate a block of raw space with `u3a_slab()`, then chop off
1233the end with `u3a_malt()` (which does the measuring itself)
1234or `u3a_mint()` in case you've measured it yourself.
1236Once again, *do not call `malloc()`* (or C++ `new`) within any
1237code that may be run within a jet. This will cause rare sporadic
1238corruption when we interrupt execution within a `malloc()`. We'd
1239just override the symbol, but `libuv` uses `malloc()` across
1240threads within its own synchronization primitives - for this to
1241work with `u3a_malloc()`, we'd have to introduce our own locks on
1242the surface-level road (which might be a viable solution).
1244### u3n: nock execution
1246The `u3n` routines execute Nock itself. On the inside, they have
1247a surprising resemblance to the spec proper (the only interesting
1248detail is how we handle tail-call elimination) and are, as one
1249would expect, quite slow. (There is no such thing as a fast tree
1252There is only one Nock, but there are lots of ways to call it.
1253(Remember that all `u3n` functions *transfer* C arguments and
1256The simplest interpreter, `u3n_nock_on(u3_noun bus, u3_noun fol)`
1257invokes Nock on `bus` (the subject) and `fol` (the formula).
1258(Why is it`[subject formula]`, not `[formula subject]`? The same
1259reason `0` is true and `1` is false.)
1261A close relative is `u3n_slam_on(u3_noun gat, u3_noun sam)`,
1262which slams a *gate* (`gat`) on a sample (`sam`). (In a normal
1263programming language which didn't talk funny and was retarded,
1264`u3n_slam_on()` would call a function on an argument.) We could
1265write it most simply as:
1268 u3n_slam_on(u3_noun gat, u3_noun sam)
1270 u3_noun pro = u3n_nock_on
1271 (u3nc(u3k(u3h(gat)),
1272 u3nc(sam, u3k(u3t(u3t(gat))))),
1278Simpler is `u3n_kick_on(u3_noun gat)`, which slams a gate (or,
1279more generally, a *trap* - because sample structure is not even
1280needed here) without changing its sample:
1283 u3n_kick_on(u3_noun gat, u3_noun sam)
1285 return u3n_nock_on(gat, u3k(u3h(gat)));
1288The `_on` functions in `u3n` are all defined as pure Nock. But
1289actually, even though we say we don't extend Nock, we do. But we
1292Note that `u3` has a well-developed error handling system -
1293`u3m_bail()` to throw an exception, `u3m_soft_*` to catch one.
1294But Nock has no exception model at all. That's okay - all it
1295means if that if an `_on` function bails, the exception is an
1296exception in the caller.
1298However, `u3`'s exception handling happens to match a convenient
1299virtual super-Nock in `hoon.hoon`, the infamous `++mock`. Of
1300course, Nock is slow, and `mock` is Nock in Nock, so it is
1301(logically) super-slow. Then again, so is decrement.
1303With the power of `u3`, we nest arbitrary layers of `mock`
1304without any particular performance cost. Moreover, we simply
1305treat Nock proper as a special case of `mock`. (More precisely,
1306the internal VM loop is `++mink` and the error compiler is
1307`++mook`. But we call the whole sandbox system `mock`.)
1309The nice thing about `mock` functions is that (by executing
1310within `u3m_soft_run()`, which as you may recall uses a nested
1311road) they provide both exceptions and the namespace operator -
1312`.^` in Hoon, which becomes operator `11` in `mock`.
1314`11` requires a namespace function, or `fly`, which produces a
1315`++unit` - `~` (`0`) for no binding, or `[0 value]`. The sample
1316to a `fly` is a `++path`, just a list of text `span`.
1318`mock` functions produce a `++toon`. Fully elaborated:
1320 ++ noun ,* :: any noun
1321 ++ path (list ,@ta) :: namespace path
1322 ++ span ,@ta :: text-atom (ASCII)
1323 ++ toon $% [%0 p=noun] :: success
1324 [%1 p=(list path)] :: blocking paths
1325 [%2 p=(list tank)] :: stack trace
1327 ++ tank :: printable
1328 $% [%leaf p=tape] :: flat text
1329 $: %palm :: backstep list
1330 p=[p=tape q=tape r=tape s=tape] :: mid cap open close
1331 q=(list tank) :: contents
1333 $: %rose :: straight list
1334 p=[p=tape q=tape r=tape] :: mid open close
1335 q=(list tank) :: contents
1339(Note that `tank` is overdesigned and due for replacement.)
1341What does a `toon` mean? Either your computation succeded (`[0
1342noun]`, or could not finish because it blocked on one or more
1343global paths (`[1 (list path)]`), or it exited with a stack trace
1346Note that of all the `u3` exceptions, only `%exit` is produced
1347deterministically by the Nock definition. Therefore, only
1348`%exit` produces a `2` result. Any other argument to
1349`u3m_bail()` will unwind the virtualization stack all the way to
1350the top - or to be more exact, to `u3m_soft_top()`.
1352In any case, the simplest `mock` functions are `u3n_nock_un()`
1353and `u3n_slam_un()`. These provide exception control without
1354any namespace change, as you can see by the code:
1356 /* u3n_nock_un(): produce .*(bus fol), as ++toon.
1359 u3n_nock_un(u3_noun bus, u3_noun fol)
1361 u3_noun fly = u3nt(u3nt(11, 0, 6), 0, 0); // |=(a=* .^(a))
1363 return u3n_nock_in(fly, bus, fol);
1366 /* u3n_slam_un(): produce (gat sam), as ++toon.
1369 u3n_slam_un(u3_noun gat, u3_noun sam)
1371 u3_noun fly = u3nt(u3nt(11, 0, 6), 0, 0); // |=(a=* .^(a))
1373 return u3n_slam_in(fly, gat, sam);
1376The `fly` is added as the first argument to `u3n_nock_in()` and
1377`u3n_slam_in()`. Of course, logically, `fly` executes in the
1378caller's exception layer. (Maintaining this illusion is slightly
1379nontrivial.) Finally, `u3n_nock_an()` is a sandbox with a null
1384The only `u3e` function you should need to call is `u3e_save()`,
1385which saves the loom. As it can be restored on any platform,
1386please make sure you don't have any state in the loom that is
1387bound to your process or architecture - except for exceptions
1388like the warm jet state, which is actively purged on reboot.
1390### u3r: reading nouns (weak)
1392As befits accessors they don't make anything, `u3r` noun reading
1393functions always retain their arguments and their returns. They
1394never bail; rather, when they don't work, they return a `u3_weak`
1397Most of these functions are straightforward and do only what
1398their comments say. A few are interesting enough to discuss.
1400`u3r_at()` is the familiar tree fragment function, `/` from the
1401Nock spec. For taking complex nouns apart, `u3r_mean()` is a
1402relatively funky way of deconstructing nouns with a varargs list
1403of `axis`, `u3_noun *`. For cells, triples, etc, decompose with
1404`u3r_cell()`, `u3r_trel()`, etc. For the tagged equivalents, use
1405`u3r_pq()` and friends.
1407`u3r_sing(u3_noun a, u3_noun b)` (true if `a` and `b` are a
1408*single* noun) are interesting because it uses mugs to help it
1409out. Clearly, different nouns may have the same mug, but the
1410same nouns cannot have a different mug. It's important to
1411understand the performance characteristics of `u3r_sing()`:
1412the worst possible case is a comparison of duplicate nouns,
1413which have the same value but were created separately. In this
1414case, the tree is traversed
1416`u3r_sung()` is a deeply funky and frightening version of
1417`u3r_sing()` that unifies pointers to the duplicate nouns it
1418finds, freeing the second copy. Obviously, do not use
1419`u3r_sung()` when you have live, but not reference counted, noun
1420references from C - if they match a noun with a refcount of 1
1421that gets freed, bad things happen.
1423It's important to remember that `u3r_mug()`, which produces a
142431-bit, nonzero insecure hash, uses the `mug_w` slot in any boxed
1425noun as a lazy cache. There are a number of variants of
1426`u3r_mug()` that can get you out of building unneeded nouns.
1428### u3x: reading nouns (bail)
1430`u3x` functions are like `u3r` functions, but instead of
1431returning `u3_none` when (for instance) we try to take the head
1432of an atom, they bail with `%exit`. In other words, they do what
1433the same operation would do in Nock.
1435### u3h: hash tables.
1437We can of course use the Hoon `map` structure as an associative
1438array. This is a balanced treap and reasonably fast. However,
1439it's considerably inferior to a custom structure like an HAMT
1440(hash array-mapped trie). We use `u3_post` to allocate HAMT
1441structures on the loom.
1443(Our HAMT implements the classic Bagwell algorithm which depends
1444on the `gcc` standard directive `__builtin_popcount()`. On a CPU
1445which doesn't support popcount or an equivalent instruction, some
1446other design would probably be preferable.)
1448There's no particular rocket science in the API. `u3h_new()`
1449creates a hashtable; `u3h_free()` destroys one; `u3h_put()`
1450inserts, `u3h_get()` retrieves. You can transform values in a
1451hashtable with `u3h_walk()`.
1453The only funky function is `u3h_gut()`, which unifies keys with
1454`u3r_sung()`. As with all cases of `u3r_sung()`, this must be
1455used with extreme caution.
1459Connected to the `~+` rune in Hoon, via the Nock `%memo` hint,
1460the memoization facility is a general-purpose cache.
1462(It's also used for partial memoization - a feature that'll
1463probably be removed, in which conservative worklist algorithms
1464(which would otherwise be exponential) memoize everything in the
1465subject *except* the worklist. This is used heavily in the Hoon
1466compiler jets (j/f/*.c). Unfortunately, it's probably not
1467possible to make this work perfectly in that it can't be abused
1468to violate Nock, so we'll probably remove it at a later date,
1469instead making `++ut` keep its own monadic cache.)
1471Each `u3z` function comes with a `c3_m` mote which disambiguates
1472the function mapping key to value. For Nock itself, use 0. For
1473extra speed, small tuples are split out in C; thus, find with
1475 u3_weak u3z_find(c3_m, u3_noun);
1476 u3_weak u3z_find_2(c3_m, u3_noun, u3_noun);
1477 u3_weak u3z_find_3(c3_m, u3_noun, u3_noun, u3_noun);
1478 u3_weak u3z_find_4(c3_m, u3_noun, u3_noun, u3_noun, u3_noun);
1482 u3_noun u3z_save(c3_m, u3_noun, u3_noun);
1483 u3_noun u3z_save_2(c3_m, u3_noun, u3_noun, u3_noun);
1484 u3_noun u3z_save_3(c3_m, u3_noun, u3_noun, u3_noun, u3_noun);
1485 u3_noun u3z_save_4(c3_m, u3_noun, u3_noun, u3_noun, u3_noun, u3_noun);
1487where the value is the last argument. To eliminate duplicate
1493`u3z` functions retain keys and transfer values.
1495The `u3z` cache, built on `u3h` hashes, is part of the current
1496road, and goes away when it goes away. (In future, we may wish
1497to promote keys/values which outlive the road, as we do with jet
1498state.) There is no cache reclamation at present, so be careful.
1500### u3t: tracing and profiling.
1504### u3v: the Arvo kernel
1506An Arvo kernel - or at least, a core that compiles with the Arvo
1507interface - is part of the global `u3` state. What is an Arvo
1508core? Slightly pseudocoded:
1512 ++ come |= [yen=@ ova=(list ovum) nyf=pone] :: 11
1513 ^- [(list ovum) _+>]
1515 ++ keep |= [now=@da hap=path] :: 4
1518 ++ load |= [yen=@ ova=(list ovum) nyf=pane] :: 86
1519 ^- [(list ovum) _+>]
1521 ++ peek |= [now=@da path] :: 87
1524 ++ poke |= [now=@da ovo=ovum] :: 42
1525 ^- [(list ovum) _+>]
1527 ++ wish |= txt=@ta :: 20
1531 ++ card ,[p=@tas q=*] :: typeless card
1532 ++ ovum ,[p=wire q=card] :: Arvo event
1533 ++ wire path :: event cause
1535This is the Arvo ABI in a very real sense. Arvo is a core with
1536these six arms. To use these arms, we hardcode the axis of the
1537formula (`11`, `4`, `86`, etc) into the C code that calls Arvo,
1538because otherwise we'd need type metadata - which we can get, by
1541It's important to understand the Arvo event/action structure, or
1542`++ovum`. An `ovum` is a `card`, which is any `[term noun]`
1543cell, and a `++wire`, a `path` which indicates the location of
1544the event. At the Unix level, the `wire` corresponds to a system
1545module or context. For input events, this is the module that
1546caused the event; for output actions, it's the module that
1549`++poke` sends Arvo an event `ovum`, producing a cell of action
1550ova and a new Arvo core.
1552`++peek` dereferences the Arvo namespace. It takes a date and a
1553key, and produces `~` (`0`) or `[~ value]`.
1555`++keep` asks Arvo the next time it wants to be woken up, for the
1556given `wire`. (This input will probably be eliminated in favor
1557of a single global timer.)
1559`++wish` compiles a string of Hoon source. While just a
1560convenience, it's a very convenient convenience.
1562`++come` and `++load` are used by Arvo to reset itself (more
1563precisely, to shift the Arvo state from an old kernel to a new
1564one); there is no need to call them from C.
1566Now that we understand the Arvo kernel interface, let's look at
1567the `u3v` API. As usual, all the functions in `u3v` are
1568commented, but unfortunately it's hard to describe this API as
1569clean at present. The problem is that `u3v` remains design
1570coupled to the old `vere` event handling code written for `u2`.
1571But let's describe the functions you should be calling, assuming
1572you're not writing the next event system. There are only two.
1574`u3v_wish(str_c)` wraps the `++wish` functionality in a cache
1575(which is read-only unless you're on the surface road).
1577`u3v_do()` uses `wish` to provide a convenient interface for
1578calling Hoon kernel functions by name. Even more conveniently,
1579we tend to call `u3v_do()` with these convenient aliases:
1581 #define u3do(txt_c, arg) u3v_do(txt_c, arg)
1582 #define u3dc(txt_c, a, b) u3v_do(txt_c, u3nc(a, b))
1583 #define u3dt(txt_c, a, b, c) u3v_do(txt_c, u3nt(a, b, c))
1584 #define u3dq(txt_c, a, b, c, d) u3v_do(txt_c, u3nt(a, b, c, d))
keysesc to close
- j k
- move down, up
- enter
- open
- esc
- back, or up a level
- 1 2 3 4
- readme, files, commits, refs
- /
- filter
- y
- copy the clone command
- c
- compare the chosen commit with another (or shift-click both)
- ?
- show or hide this list